SkillFlow: Flow-Driven Recursive Skill Evolution for Agentic Orchestration

Mingda Zhang, Tiesunlong Shen, Haoran Luo, Wenjin Liu, Zikai Xiao, Erik Cambria, Xiaoying Tang

arXiv:2605.14089 · 2026-05-16 공개 · arXiv · PDF

credit-assignment skill-evolution reward-maximization flow-based agentic-orchestration ttb-loss llm-supervisor task-orchestration

Abstract

In recent years, a variety of powerful LLM-based agentic systems have been applied to automate complex tasks through task orchestration. However, existing orchestration methods still face key challenges, including strategy collapse under reward maximization, high gradient variance with opaque credit assignment, and unguided skill evolution whose decisions are typically made by directly prompting an LLM to judge rather than derived from principled training signals. To address these challenges, we propose SkillFlow, a flow-based framework that takes a trainable Supervisor as the agent and a structured environment with dynamic skill library and frozen executor, automating task orchestration through multi-turn interaction. SkillFlow employs Tempered Trajectory Balance (TTB), a regression-based flow-matching loss that samples trajectories proportional to reward, preserving diverse orchestration strategies rather than collapsing to a single mode. The same flow objective yields a jointly learned backward policy that provides transparent per-step credit assignment at zero additional inference cost. Building on these flow diagnostics, a recursive skill evolution mechanism determines when to evolve, what skills to create or prune, and where decision gaps lie -- closing the loop from training signal to autonomous capability growth. Experimental results on 14 datasets show that SkillFlow significantly outperforms baselines across question answering, mathematical reasoning, code generation, and real-world interactive decision making tasks. Our code is available at https://anonymous.4open.science/r/SkillFlow-E850.

한국어 요약

📋 한 줄 요약

**[에이전트 오케스트레이션 / 흐름 매칭 RL]** 보상에 비례한 trajectory 샘플링과 재귀적 스킬 진화를 결합해 에이전트 오케스트레이션을 종단간 학습하는 흐름 기반 프레임워크 SkillFlow 제안.

🎯 핵심 기여도

💡 핵심 아이디어

오케스트레이션을 단일 최적 경로가 아니라 보상 분포에 비례하는 trajectory 분포 학습으로 본다. 그러면 다양한 전략이 자연 보존되고, 학습된 backward 흐름이 곧 credit assignment 신호가 된다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: RL 기반 에이전트 오케스트레이션에 GFlowNet의 다양성 보존·credit assignment 이점을 가져와 자율 능력 성장 루프를 닫음. **한계**: 흐름 매칭 학습은 정상 분포·온도 스케일링에 민감, 매우 큰 skill library에서 진화 비용이 증가할 수 있음.

🚀 실용적 활용